University of Illinois at Chicago
University of Illinois at Chicago: UIC INDIGO (INtellectual property in DIGital form available online in an Open environment)Not a member yet
21439 research outputs found
Sort by
A Framework for Enhancing Graph Neural Networks Using Explanations
No full textMany real-world problems are modeled as graphs that represent relationships between entities. Graph Neural Networks (GNNs) are a powerful variant of neural networks that combine vertex and edge attributes with node neighborhood structures to infer properties of graph data. Message Passing Neural Networks (MPNNs), a common type of GNN, leverage the expressiveness of the first-order Weisfeiler-Leman (1-WL) algorithm for learning representations for classification tasks. However, 1-WL has known limitations in expressiveness and these limitations pose serious limitations on GNN performance. Separately, eXplainable Artificial Intelligence (XAI) is a sub-field of Machine Learning focused on addressing the “black-box” nature of neural networks. Several projects, such as GNNExplainer, have explored providing post-hoc explanations for GNN predictions. This current work combines XAI methods with graph mining to develop a computational framework to improve GNN performance.
The following are the main themes of this work: (1) A new computational framework called Explanation Enhanced Graph Learning (EEGL) to address the performance limitations of GNNs. We achieve this by annotating the input with relevant local structural information based on explanation artifacts and graph mining. Through experiments, we show that data annotated in this way results in higher model performance. (2) We study four different types of noise in our synthetic data and their effects on GNN learnability. We then show that EEGL can mitigate these adverse effects leading to improved performance even in noisy data. (3) GNNs as Logical Classifiers: Logical characterization involves a structured way to analyze and define the expressiveness of GNN models, e.g., “learning a query,” meaning learning a node classification problem in a unified manner across all graphs using a single logic formula. Through experiments, we examine the inductive learning characteristics of GNNs and the models’ ability to generalize on graphs that encode the same logical rule across structurally diverse graphs. (4) Philosophy of Science Perspective on Explainable AI: We briefly explore the philosophy of science perspective on Explainable AI. In this study, we discuss a high-level framework called ExpSpec for contextualizing requirements and defining a set of requirements for explanations
A Bending Norm Criterion for Coarsely Embedded Pleated Planes
No full textGiven a surface, , of negative Euler characteristic and some maximal lamination, , Bonahon developed shear-bend coordinates, which identify an open subset of the character variety, \chi(\pi_1(S), \mrm{PSL}_2\C), whose images contain the quasi-Fuchsian representations with an open subset of a finite dimensional \C-vector space, \mc{H}(\lambda, \C). We show that at every Fuchsian representation, there is some definite radius depending only the choice of a train track (which yields a norm on \mc{H}(\lambda,\C)) and on the injectivity radius determined by the Fuchsian representation so that the ball of this radius in the bend coordinates is entirely contained in the quasi-Fuchsian locus in \chi(\pi_1(S), \mrm{PSL}_2\C). This work is foreshadowed by the work of Epstein, Marden, and Markovich in which they prove a similar result in the special case that the bending cocycle is also a transverse measure
The Role of Inclusion-Focused Teacher Training in Promoting Inclusive Classroom Practices
No full textChildren with disabilities are suspended and expelled at disproportionately high rates (Zeng et al., 2020), as are children who are living with mental health support needs (Zeng et al., 2019). In an early childhood classroom, unmet needs can manifest in behavioral challenges, such as acting out, withdrawal, or disruptions in the classroom (Gleason, P.M., 2017). Without training or support in understanding these behaviors, teachers may resort to punitive measures, supporting a cycle where children are punished or removed from a classroom when they most benefit from individualized instruction (Zinsser et al., 2022). When teachers successfully engage in inclusive practices, children with disabilities and mental health needs can benefit academically and socially from early educational experience (Buysse et al., 1999).
Despite the importance of these inclusive and supportive teaching practices, few teacher preparation programs provide teachers with explicit training in this area, which may contribute to disparities in discipline. The current study seeks to evaluate the effectiveness of a 12-week inclusive teaching course, part of an alternative licensure program. Teachers working in high-need Chicago preschools (N=57) across two cohorts completed the course. They were primarily women (89.5%) of color (61.4% Black or African American, 24.6% Latina, 3.5% Asian) and had, on average, 15.2 years of prior early childhood teaching experience.
This paper will include analyses of data from three timepoints: baseline, pre-Inclusion semester, and post-Inclusion semester. The pre-post surveys included 45 Likert-style questions capturing participants' agreement with statements in five areas related to inclusion knowledge, confidence, and practices. One year before the Inclusion semester (at admission into the program), participants completed a baseline shortened version of the pre-post survey (referenced herein as the Baseline Inclusion Survey), allowing us to assess change before and during the focal semester. Final analyses using the full sample will also test for stability in teachers’ beliefs and practices prior to the start of the inclusion-focused semester to enhance our causal inference.
This study will demonstrate the potential impact of directed training for teachers on inclusive practices at the pre-licensure level. By equipping teachers with the skills, knowledge, and confidence to support and advocate for the children in their classroom with disabilities, delays, and unmet mental health and behavioral needs, programs such as this can potentially disrupt the preschool-to-prison pipeline that disproportionately affects children with disabilities
Acoustic Emission-Based Structural Health Monitoring Framework for High Temperature Piping Systems
No full textWave-based Nondestructive Evaluation (NDE) of metallic piping systems in high-temperature environments faces challenges such as sensor temperature limitation, changes in wave characteristics due to temperature-induced material property, fluid-structure interactions, and geometric complexities including pipe elbows. The common form of damage that occurs in such critical piping systems is creep, which is the time-dependent deformation of metallic materials under sustained stress at high temperatures. Detecting and monitoring creep damage is essential for maintaining structural integrity. Acoustic Emission (AE) is an NDE method that detects propagating elastic waves released by active flaws in solids and enables real-time monitoring.
This research provides a structural health monitoring (SHM) framework for monitoring and detecting creep damage in piping systems under high temperatures using AE, studied both numerically and experimentally through tests on metallic structures ranging from small dog-bone samples to a large-scale piping system. This research has four main outcomes: (1) Linear regression-based algorithm for detecting creep stage transition using AE data such as localized AE events, which can be adaptable to metallic materials under various experimental variables such as temperature and stress levels. It bridges the gap between traditional creep strain analysis and real-time damage monitoring. (2) Numerical models for simulating and visualizing wave propagation, analyzing geometric effects, and investigating multi-physics interactions coupled with piezoelectric sensor for real-world signal response simulation. (3) Implementing multi-task learning (MTL) for improving AE source localization accuracy in pipe structure with elbows. (4) Developing a regression-based location mapping approach based on AE system connectivity to enable source localization when the AE source is outside the sensor network in complex piping systems
The Influence of Population Variation and Obesity on Facial Growth
No full textMany studies have indicated that childhood obesity can affect the timing of dental development, accelerating dental maturation (e.g., Hilgers et al., 2006; Mack et al., 2013; Nicholas et al., 2018; Must A et al., 2012). Prior work has also suggested that children with obesity may also experience precocious facial growth and perhaps differences in mandibular dimensions (Vora et al., 2022) especially condylion-pogonion (Ohrn et al., 2002; Sadeghianrizi et al., 2005; Gordon et al. 2021). What remains unclear is whether this difference in timing varies by ancestry group and whether there is an interaction between demographic characteristics and obesity. Variations in facial growth patterns (shape) are also less clear, with studies showing inconsistencies in which aspects of facial shape may be affected (e.g., the contrasting results of Vora et al. 2022 and Gordon et al. 2021).
The primary aim of this study was to examine the interaction between demographic characteristics (e.g. ancestry, sex) and environmental factors such as BMI, SES, and food insecurity on the timing and pattern of facial skeletal growth. We hypothesize that (1) children with obesity will show relatively accelerated facial growth regardless of self-reported ancestry; (2) that variation in facial shape related to obesity will be mediated by self-reported ancestry.
In this study, we examined facial growth in a sample of contemporary orthodontic patients (n=105) who were prospectively recruited during their “records” (initial) appointment at the UIC Department of Orthodontics. These subjects had lateral cephalograms and height/weight taken during their appointment. They were also given a detailed 12-item demographics questionnaire. The questionnaire consisted of questions regarding self-identified ancestry/ethnicity; ancestry/ethnicity of grandparents; household income; parental educational attainment; and food insecurity.
Geometric morphometric analysis of facial shape was carried out. Descriptive statistics and linear regression were used for data analysis. Of the 105 patients recruited, 97 children and adolescents (median age 13.2 years, IQR 11.5-15 years) met the inclusion criteria for this study. No significant association was found between BMI and facial form. However, principal component 3 (PC3, 9.87% of total variation) displayed significance by age (p=0.001), sex (p=0.013), non-Hispanic Asian (p=0.048), and Black self-identified ancestry (p< 0.001) while PC4 (5.36% of total variation) was significant by age (p=0.024). Non-Hispanic Asians had lower PC3 scores, characterized by relatively more obtuse gonial angles, greater chin projection, and less subnasal alveolar prognathism. In contrast, Black subjects were more likely to have higher PC3 scores, characterized by greater subnasal alveolar prognathism and more acute gonial angles. Older subjects had greater nasal bridge protrusion, ramus length, and midfacial protrusion.
This study failed to reject our null hypotheses, suggesting no significant differences in the timing or pattern of facial growth between children with and without obesity in our sample. Due to our failure to identify an association between obesity and facial shape in our sample, we were unable to evaluate whether this pattern was mediated by self-reported ancestry (Aim 2). Given the relatively small sample size and contrasting results with prior published work, further research is needed in this area to resolve potential associations between BMI and facial growth. Differences in facial growth patterns may influence orthodontic treatment needs and the optimal timing for orthodontic interventions
Surrogate-Based Adaptive Algorithms for Optimizing Design in Complex Expensive Black-Box Systems
No full textOptimization of complex systems and processes is critical across various fields such as engineering, finance, and healthcare, where it significantly enhances system performance, reduces costs, and ensures overall success. Specifically, global optimization of an unknown, complex, and expensive objective function, commonly referred to as a ``black-box function'', poses a significant challenge. In a black-box system, the internal mechanisms are not visible or understood; only the inputs and outputs can be observed. This lack of transparency complicates the optimization process because it relies on outcomes without an understanding of the underlying processes. Traditional optimization methods often fall short in these scenarios, where the characteristics of the objective function are unknown. Consequently, evaluating these systems typically requires numerous costly simulations or experiments, making the processes both time-consuming and computationally demanding.
Surrogate Optimization (SO) is a prevalent and promising technique for black-box optimization which employs a low-cost surrogate model to guide the optimization process toward the global optimum. However, SO faces significant challenges, particularly in managing the exploration-exploitation trade-off. This trade-off requires the surrogate model to balance between exploring new regions of the function space to uncover potentially superior solutions (exploration) and exploiting regions that are currently believed to be promising based on the surrogate's predictions (exploitation).
Moreover, surrogate models often struggle to accurately represent the behavior of the function in high-dimensional spaces, making optimization more difficult and computationally intensive. These challenges are amplified in batch evaluation settings, where black-box function evaluations occur in batches rather than sequentially. This setup adds complexity to balancing exploration and exploitation, as decisions must consider multiple function evaluations simultaneously.
In batch SO, the challenge of promoting diversity while pursuing optimal solutions is paramount, hindering the discovery of promising solutions and reducing efficiency. Addressing these challenges requires innovative strategies to effectively manage the exploration-exploitation trade-off and improve the performance of SO in complex optimization scenarios.
Explainability is essential for gaining user trust in advanced optimization methods, particularly black box optimization. By providing clear, interpretable explanations of how an algorithm makes its decisions, stakeholders gain a better understanding of the process and are more likely to adopt and rely on these techniques. This transparency not only strengthens confidence in the outcomes but also makes it easier to identify potential biases or flaws. In SO, however, the complexity of both surrogate models and sampling strategies (for example, acquisition functions) often leads to a lack of clarity. While existing research has largely focused on improving convergence to global optima, the practical explainability of newly proposed strategies, particularly in batch evaluation settings, remains insufficiently explored.
This thesis introduces novel approaches for balancing the exploration-exploitation trade-off by prioritizing diversity for batch sampling. These strategies prioritize diverse candidate batch generation through adaptive sampling techniques, infusing vitality into the optimization process and effectively exploring uncharted regions of the search space.
Empirical validation demonstrates that these methods effectively navigate complex design landscapes, as shown by testing on diverse real-world benchmark problems, including DNA binding, Airfoil design, and MNIST hyperparameter optimization. Beyond theoretical advancements and empirical validation, this thesis lays the groundwork for a paradigm shift, empowering practitioners to approach complex optimization challenges with renewed precision by promoting diversity and elevated exploration. Additionally, the thesis explores the impact of SO's algorithmic hyperparameters on the exploration-exploitation trade-off to establish a robust framework. It aims to present a holistic view of surrogate optimization as a cohesive system, offering fresh insights into learning and optimizing hyperparameters across different frameworks without the need for manual tuning or incurring significant computational costs.
Lastly, this thesis addresses the primary challenge of explainability in Surrogate Optimization by introducing a comprehensive, model-agnostic framework of explainability metrics that can potentially enhance user trust. These metrics provide intermediate and post-hoc explanations to guide practitioners in understanding the SO process before and after costly evaluations. To evaluate the impact of these metrics in real-world scenarios, we applied them to benchmarks such as Robot Pushing and rover trajectory challenges, observing how deeper insights can enhance user confidence in the optimization process
Towards Intelligent Text-Based Agents with Deep Learning
No full textNowadays, intelligent agents have gained a lot of attention and popularity due to the fast development of computing techniques and resources. Text-based intelligent agents can assist people in processing a vast amount of text information in their work and life. Several requirements for such agents need to be satisfied to effectively help people improve productivity, including understanding a single piece of text correctly, multi-document understanding, extracting/generating key concepts of a piece of text, and summarizing a long document. In this presentation, four research topics are discussed to help build those abilities for the intelligent text-based agents with deep learning methods. Firstly, I propose a neural network-based method utilizing information maximization to enhance the single-document understanding of the agent. Secondly, I design a hierarchical bi-directional self-attention network for multi-document understanding. Thirdly, I propose a joint learning method to generate key attribute values for a product from its textual title and description. Fourthly, I design a two-stage pipeline method to generate multiple aspect-based summaries for long meeting transcripts. In these research works, I only use public datasets including RCV1-V2, WOS, OpenReview, PeerRead, MEPAVE, AMI corpus and ICSI corpus
Uncovering the Role of Risk Factors in Patency of Cerebrovascular Bypass
No full textIntroduction: Extracranial-intracranial (EC-IC) bypass has been well described in chronic vaso-occlusive cerebrovascular diseases, such as Moyamoya disease (MMD) and atherosclerotic disease (AD), and in the treatment of complex aneurysms and tumors. In general, demographics and comorbid conditions play a role in vascular health, yet their specific impact on cerebrovascular bypass patency remains unclear. This study examines disease etiology and other modifiable and non-modifiable patient-specific risk factors as potential contributing factors to bypass failure.
Methods: An institutional database from the University of Illinois Chicago with intracranial bypass procedures between 08/2001-05/2022 was retrospectively reviewed. Patients with bypass for all causes (e.g., aneurysm, atherosclerotic disease, Moyamoya disease) were included. Data on baseline patient demographics and medical history, surgical technique, and both intraoperative and post-operative flow-related measurements were collected. Comparisons between two steno-occlusive disease states were first performed, followed by comparisons between male and female sexes.
Results: In the first analysis, 232 patients met inclusion criteria (AD n=108; MMD n=124). Average age and sex significantly differed between groups (AD 57.2 years, 56.5% male; MMD 36.6 years, 31.5% male, p<0.001). Modified Rankin scale (mRS) at surgery and at follow-up were higher in the AD group, p=0.004 and <0.001, showing a slightly worse baseline functional status, and higher rates of stroke were observed in the AD group by last follow-up (p=0.005). At last follow-up, rates of occlusion did not differ between the AD and MMD groups (25.2% vs. 25.4%, respectively). Of occluded bypasses, the AD group had more bypasses occluded within 1-week compared to MMD (51.9% vs. 34.4%, p=0.176), although not statistically significant. In patients with more than 1-year follow-up and more than 2-year follow-up, MMD tended to have higher rates of occlusion (31.2% vs. 26.1%, p=0.558, and 26.4% vs. 20.7%, p=0.564), though again these differences were not statistically significant. Flow measurements did not differ between AD and MMD, but in subgroup analyses of patients with AD and with MMD, both bypass flow and cut flow index predicted occlusion in both groups.
In the second analysis, all 357 patients within the database were considered, with 141 male (39.5%) and 216 female (60.5%) with average age 49.0+/-16.7 years and average follow-up 1.97 years. Bypass patency at last follow-up was 84.4% (n=114) for men vs. 69.2% (n=148) for women (p=0.001). Differences were seen in underlying diagnoses, with more aneurysm and Moyamoya cases represented in female sex (p<0.001); irrespective of diagnosis, lower patency rates were seen in women when considering bypass for aneurysm (p=0.032), Moyamoya disease (p=0.035), and for atherosclerotic disease (p=0.159). Medical comorbidities were seen at higher rates in men, with comorbidity score 2.7 vs. 2.1 (p<0.001). Cut flow was higher in men 59.2 vs. 51.1 (p=0.028), with no differences in intraoperative bypass flow, cut flow index (CFI), or follow-up quantitative magnetic resonance angiography (QMRA). Propensity score-matched analysis found females have a 2.71 higher chance of bypass occlusion after adjusting for CFI (p=0.017, 95% CI: 1.19-6.18).
Conclusion: Despite different etiologies for bypass, rates of occlusion at last follow-up did not vary between different steno-occlusive disease groups, although short-term follow-up would suggest earlier bypass failure in AD and extended follow-up trended toward higher occlusion rates in MMD. Additionally, patients with AD were more likely to have further stroke by last follow-up. Importantly, the bypass flow and cut flow index at the time of surgery predicted occlusion in both AD and in MMD. Regarding the influence of sex on bypass patency, women were less likely to have patent bypasses at last follow-up, despite having less medical comorbidities than men and despite having similar intraoperative and perioperative flows
The Effects of Hypoxia on Heart and Brain Tissue of the African Naked Mole Rat
No full textThis study investigates the impact of hypoxia on mammals, emphasizing the adaptive mechanisms that sustain cellular equilibrium under oxygen-deficient conditions. Mitochondrial oxidative phosphorylation is highlighted as the primary and efficient method for energy production in aerobic scenarios, in contrast to the less efficient glycolysis pathway.
Using the naked mole-rat as a model organism, the research examines its exceptional adaptations to hypoxia, which enable survival in low-oxygen environments for at least 18 minutes. This resilience is demonstrated through brain slice models that maintain synaptic function under extreme oxygen deprivation. Neural adaptations are further explored using EEG recordings, lactate measurements, and NMDA infusions into the caudate, evaluating brain activity during wakefulness, sleep, hypoxia, and recovery, and investigating lactate's role in brain metabolism and neuroprotection.
Additionally, the study includes Mexican free-tailed bats and neonate mice to examine hypoxia tolerance across a broader range of species, providing insights into the mechanisms of hypoxia resilience in mammals
Familismo in the Academy: Challenges and Strategies among Latina Graduate and Professional Students
No full textLatina women have outpaced their Latino male counterparts in holding advanced degrees since 2010 (Mora and Lopez 2023). It is evident that although Latina students are making significant strides in increasing the educational attainment in their ethnic group, they are still severely underrepresented when compared to their counterparts across racial and gender groups. Through 64 semi-structured interviews with Latinas in graduate and professional programs across the U.S., I explore the challenges and strategies that Latina students face and implement as they navigate academia. I find that Latina students experience a familismo paradox that describes the phenomenon that although Latina students often face contradictions from their families, the values instilled in them provide unique support. Latina students also often come from families with lower levels of education and income and lack access to educational resources. Still, their academic outcomes are making strides comparable to white students. I also find that Latina students apply these familismo values to the academy and on social media to build community and refashion their identities. Lastly, I present three recommendations that Latina students suggest for higher education institutions to support them better, such as increasing funding opportunities, hiring faculty of color, and creating efforts that integrate their families. Latina students are entering advanced degree programs more than ever, so it is important for institutions to meet their particular needs
Data sources:
Access Repository Dashboard
Do you manage Open Research Online? Become a CORE Member to access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard! 👇